Strojniški vestnik - Journal of Mechanical Engineering 60(2014)2, 124-134 Received for review: 2013-05-17

© 2014 Journal of Mechanical Engineering. All rights reserved. Received revised form: 2013-09-20
DOI:10.5545/sv-jme.2013.1222 Original Scientific Paper Accepted for publication: 2013-12-10

Improvement of Efficiency Prediction for a Kaplan Turbine

with Advanced Turbulence Models
Jošt, D. – Škerlavaj, A. – Lipej, A.
Dragica Jošt1,* – Aljaž Škerlavaj1 – Andrej Lipej1
1Turboinštitut, Slovenia

A comparison between numerical simulations and measurements of a six-blade Kaplan turbine is presented in order to determine an
appropriate numerical setup for accurate and reliable simulations of Kaplan turbines. Values of discharge, torque and losses obtained by
different turbulence models are compared to each other and to the measurements. Steady state simulations with various turbulence models
tend to predict large errors at full discharge rate, which are the result of underestimated torque on the shaft and overestimated flow energy
losses in the draft tube. The results were slightly improved with the curvature correction (CC) and Kato-Launder (KL) limiter of turbulence
production. Transient simulations were performed with shear-stress-transport (SST) turbulence model, the scale-adaptive-simulation (SAS) SST
model, and with zonal large-eddy-simulation (ZLES). Details about turbulent structures in the draft tube are illustrated in order to explain the
reasons for differences in flow energy losses obtained by different turbulence models. The effects of advection schemes and mesh refinement
were tested. It was shown that all of the transient simulations considerably improved results at full discharge rate. The largest improvement
was achieved with the SAS SST and the ZLES models in combination with the bounded central differential scheme. In addition, it was shown
that the ZLES model produced accurate results at all operating points, with discrepancy lower than 1%.
Keywords: water turbine, Kaplan turbine, efficiency prediction, CFD, turbulence models, ZLES

0 INTRODUCTION Another compromise is to perform steady-state

simulations instead of transient ones. Flow in turbines
For more than 20 years, computational fluid dynamics is unsteady and usually a converged steady-state
(CFD) has played an important role in the gradual solution cannot be obtained [2]. Besides, two-equation
improvement of the characteristics of axial water Reynolds-averaged Navier Stokes (RANS) models,
turbines. The relative effect of hydraulic shape including the shear-stress-transport (SST) model that
modifications on the efficiency curve can be quite became a standard in turbomachinery, are not able to
reliably predicted, yet the prediction of the absolute model all flow structures in the draft tube.
value of efficiency is still a challenge. The main An important attempt to determine the
problem is an inaccurate flow simulation in the draft applicability of state-of-the-art CFD simulations for a
tube. In this paper we present a case where extremely Kaplan draft tube was the Turbine 99 workshop held
poor results of steady-state analysis were improved in Porjus, Sweden, in the years 1999, 2001 and 2005.
by transient simulations and advanced turbulence The axial and swirl velocity components at the inlet
models. (with rms-values and one Reynolds’ stress component)
A simulation of the Kaplan turbine is in general and the pressure distribution around the outlet cross
more time consuming than a simulation of the Francis section were available before the workshop. For the
turbine. The Kaplan turbine is double-regulated. third meeting, besides two inlet boundary conditions
Therefore, the overall efficiency curve is an envelope (the first one with a prescribed axi-symmetrical
of partial efficiency curves, which are determined at inlet velocity profile, and the second one with a
a fixed angle of runner blades. The specifics of the measured phase-averaged resolved inlet velocity
Kaplan turbine are hub and tip clearance. At large profile) also computational grids with 1 million and
guide vane openings there is often a clearance between 2.5 million nodes with values for y+ equal to 1 and
overhanging guide vanes and the bottom ring. For 50 were provided by the organiser. Numerical results
accurate simulations it is recommended to take all obtained by different CFD codes were compared to
such details into account. However, this results in the experimental data for pressure recovery factor,
large computational grids. pressure distributions along the draft tube walls and
Often engineers have to make a compromise in a detailed velocity field in one downstream cross
order to reduce computational time. An extreme case section. One of the conclusions was that much
is to simulate only one periodic segment of runner attention must be paid to grid quality and boundary
blades and a draft tube. In [1] it is reported that such conditions [3]. Too coarse a grid and too high values
an analysis was not satisfactory, until the effect of the of y+ significantly reduced the accuracy of the pressure
spiral casing with stay and guide vane cascades on recovery factor. All tested RANS models predicted
conditions at the runner inlet, were taken into account. too weak a secondary flow [4]. Although the results
124 *Corr. Author’s Address: Turboinštitut, Rovšnikova 7, 1210 Ljubljana, Slovenia, [email protected]
 Strojniški vestnik - Journal of Mechanical Engineering 60(2014)2, 124-134

of SST were significantly better than the results of the the Wilcox k–ω model [11], the Baseline (BSL) k–ω
k–ε model [5], they did not reach the same level of model [12], the SST model [12] and [13] and the
accuracy throughout the draft tube as the LES model. ε-based SSG RSM [14]. We have used two scale-
The best results were obtained by Kurosawa and resolving simulation (SRS) models, the SAS SST and
Nakamura [6] using the dynamic Smagorinsky LES the ZLES model.
model and axi-symmetrical inlet boundary profile. The SAS SST turbulence model [15] is a so-
Since 2005, several papers with the same draft tube called second generation URANS model, according to
using measured quantities from the Turbine 99 case, classification [16]. The model is essentially the SST
have been published [7] and [8]. In [7], the results turbulence model with an additional source term QSAS
obtained with the SST model, with time dependent in the ω transport equation [17]:
angular resolved inlet velocity boundary condition   L  
on a mesh, with 6 million cells, were presented. QSAS = max  f   , 0 . (1)
The conclusion was that by using time dependent   LvK  
angular resolved inlet boundary conditions and grid
refinement, only limited improvement of numerical The term QSAS can detect the unsteadiness of
results can be achieved. A possible means of further the solution through the comparison of the RANS
improvement might be in turbulence modelling. length scale L to the von Karman length scale LvK.
In 2011, a numerical research project about The result of the unsteadiness is an increased value of
the modelling of axial turbines was carried out at QSAS, which results in decreased turbulent viscosity.
Turboinštitut [9]. Its purpose was to determine the Consequently, the SAS SST develops an LES-like
optimal setup of a numerical model, which would be solution in unsteady regions. At the same time, the
capable of reproducing the measured overall efficiency model provides standard RANS capabilities in stable
curve. The study was focused on the effect of hub and flow regions. If the time step size is too large the
tip clearance, on the mesh refinement effect, effect of unsteady structures can’t be resolved and the model
two- or seven- equation RANS turbulence models, obtains an RANS or URANS solution [18].
and type of simulation (steady-state or transient). The main idea of the ZLES model [17] is to
None of the results were satisfactory. Therefore, resolve the flow inside a predefined zone with the
it was concluded that the only way to predict LES model, and the rest of the domain with the
efficiency accurately at all operating regimes was to RANS model. In CFX, the model source term in the
use advanced turbulence models, such as the scale- k-equation forces the eddy viscosity to be equal to the
adaptive-simulation SST turbulence model (SAS LES subgrid-scale viscosity inside the user-specified
SST), and zonal large eddy simulation (ZLES). The zone. The synthetic turbulence at the RANS-LES
improvement of the results by these two turbulence boundary is based on harmonic flow generator [19]
models was presented in [2]. Such an analysis is too acting through a special source term in the momentum
time-consuming to be used in the design process, but equation. In CFX, the ZLES model is available with
at least for the final geometry it is worth doing. all k–ω turbulence models (Wilcox, BSL, SST, BSL,
In this paper the results are presented for the same explicit algebraic RSM, DES SST and SAS SST). In
Kaplan turbine as in [2]. While the paper [2] was the presented simulations, the zone of the ZLES model
focused on comparison of predicted turbine efficiency was defined within the SAS SST simulation. The zone
to the measured values, in this paper also the values started just after the interface between the runner and
of discharge, torque and losses obtained by different the draft tube, and it is extended to the outlet of the
turbulence models are compared to each other computational domain.
and to the measurements. The effects of different Some turbulence models were used in
discretisation schemes for advection term and of grid combination with curvature correction (CC) [20] and
density are also presented. with the Kato-Launder limiter (KL) [21] of turbulence
Numerical simulations were performed with the production. The CC option captures the effects of
Ansys CFX solver [10]. streamline curvature and system rotation. When the
CC option is selected, the production terms in k and ω
1 TURBULENCE MODELS AND DISCRETISATION SCHEMES transport equations are multiplied by the upwards and
downwards limited curvature correction function [20].
In this paper, several turbulence models were used. The function is defined as:
Most of them are well known RANS turbulence
models, such as the standard k–ε turbulence model,
Improvement of Efficiency Prediction for a Kaplan Turbine with Advanced Turbulence Models 125
 Strojniški vestnik - Journal of Mechanical Engineering 60(2014)2, 124-134

2r *  two vertical piers. Tip clearance was modelled while

f rot = (1 + cr1 ) 1 − cr3 tan −1 (cr2 r )  − cr1 , (2) hub clearance was neglected. The grid in the spiral
1 + r*  
casing with stay vanes was unstructured, while the
where the r* and r are functions of the strain rate grids in the other turbine parts were structured. All the
tensor Sij and of the effective rotation rate tensor grids satisfied the recommendations for orthogonality
Wij, which includes the term of the system rotation and the aspect ratio of elements. Near the walls the
Ωrot. The empirical constants cr1, cr2 and cr3 are set grids were refined to get recommended values of y+.
equal to 1, 2 and 1, respectively. The lower limit of For the draft tube and the draft tube prolongation,
the function is 0, which represents strong convex basic and refined grids were used. For the other turbine
curvature, whereas the upper limit is 1.25, which parts, our previous studies [9] showed that with proper
represents strong concave curvature. values of y+, the number of nodes used in this case
The KL limiter uses a different formulation to (see Table 1) is sufficient, and that by additional grid
deal with the excessive production of turbulence refinement only a negligible improvement of results
energy at stagnation points, than it is the default in can be obtained. The computational grid for the
CFX. In CFX, the limiter is defined as a constant, complete computational domain can be seen in Fig. 1.
multiplied by the density and the turbulence eddy
dissipation. The KL limiter replaces one of the strain
rates in the production term with the magnitude of
vorticity [10] and [21].
In the simulations, we used either ‘high resolution’
or the bounded central difference scheme. The high-
resolution scheme (HRS) is a bounded second order
upwind biased discretisation. The scheme calculates
the bounded values based on the procedure of the
Barth and Jespersen’s scheme [22] and [23]. The
bounded central difference scheme (BCDS) is based
on the normalised variable diagram and blends from Fig. 1. Computational domain and basic grid
the central difference scheme (CDS) to the first-order
upwind scheme when the convection boundedness Table 1. Number of nodes for all turbine parts (in basic grid – BG
and in fine grid – FG)
criterion [24] is violated.
For convenience, short labels are used for Turbine part Number of nodes
turbulence models and discretisation schemes: SST Semi spiral casing with stay vanes 1,480,999
for the Shear Stress Transport model, SAS for the SAS Guide vane cascade 2,755,496
SST model, ZLES for the SAS SST ZLES model. So, Runner 1,858,374
for example, ZLES BCDS means the SAS SST ZLES Draft tube (BG) 1,786,432
turbulence model with BCDS used for the advection Draft tube prolongation (BG) 398,056
term. Draft tube (FG) 6,169,935
Draft tube prolongation (FG) 1,681,992
2 NUMERICAL PREDICTION OF FLOW Total (BG) 8,279,357
AND ENERGETIC CHARACTERISTICS Total (FG) 13,946,796

In this paper the results of a detailed numerical The numerical analysis was done in three stages.
analysis of flow in a 6-blade Kaplan turbine which Firstly, a steady-state analysis at the local best
operates at middle head (ψ = 0.44) are presented. efficiency points for three angles of runner blades was
The model of the turbine was tested on a test rig in performed with several turbulence models. Secondly,
accordance with international standard IEC 60193 a transient analysis was done with three turbulence
[25], so we were able to compare the numerical results models using two discretisation schemes (SST HRS,
with the measured ones. Numerical simulations were SAS HRS, SAS BCDS and ZLES BCDS) at only
done for three angles of runner blades at constant one operating point. Finally, a transient analysis was
head. performed at several operating points for three angles
The turbine consists of semi-spiral casing with of runner blades, with the ZLES turbulence model and
two vertical piers, 11 stay vanes and a nose, 28 guide BCDS.
vanes, a 6-blade runner and an elbow draft tube with
126 Jošt, D. – Škerlavaj, A. – Lipej, A.
 Strojniški vestnik - Journal of Mechanical Engineering 60(2014)2, 124-134

2.1 Steady-State Analysis with Different Turbulence Models only analysis at the local best efficiency points for
three angles of runner blades (see Table 2) was
Steady-state analysis was performed on the basic grid performed, with guide vane opening being determined
with k–ε, k–ω, BSL, SST and SSG-RSM turbulence from experimental data.
models. For two-equation RANS models, simulations
were repeated with CC and KL limiter of production Table 2. Operating points for steady state analysis by different
term. For the k–ε and SSG RSM models, scalable turbulence models
wall functions were used. For k–ω based turbulence Operating point β [°] φ/φBEP ψ/ψBEP
models, automatic near-wall treatment was used. The OP1 12 0.64 0.86
automatic treatment allows a gradual switch between OP2 20 0.95 0.86
wall functions and the low-Reynolds number method. OP3 28 1.31 0.86
For discretisation of the advection term, the HRS
implemented in ANSYS-CFX was used. The simulations were done at constant head and
In the case of the SST CC KL and the SSG RSM rotational speed. From numerical results the values of
turbulence models, three partial efficiency curves discharge, torque on the shaft, flow energy losses in
were simulated (see Section 2.3). In the other cases, all turbine parts and turbine efficiency, were obtained.

Fig. 2. Comparison between steady-state results obtained with different turbulence models and the measured values

Improvement of Efficiency Prediction for a Kaplan Turbine with Advanced Turbulence Models 127
 Strojniški vestnik - Journal of Mechanical Engineering 60(2014)2, 124-134

The results at local best efficiency points for Calculated efficiency values at all three operating
each blade angle are presented in Fig. 2. The values points and for all turbulence models were smaller than
of discharge, torque on the shaft and efficiency were the measured ones (relative values are smaller than 1).
divided by measured values at the same operating For all two-equation models the values of efficiency
points. That means that all measured values in relative increased by using KL and CC, due to smaller flow
form are equal to 1 and are therefore not presented in losses. At OP1 and OP2 all turbulence models
the diagrams. The values of flow energy losses are predicted the efficiency values rather well. At OP1 the
presented in percentage of head. differences between measured efficiency and those
For all operating points and all turbulence obtained by SSG RSM and by SST CC KL were 0.38
models, the calculated values of discharge were larger and 1%, respectively. The same differences at OP2
than the measured ones (relative values are larger than were 0.8 and 1.43%. At OP3, the discrepancy between
1). The discrepancy was the largest when turbulence calculated and measured efficiency values was very
models k–ε, k–ε CC KL and SSG RSM were used. large, for some of the two-equation models without CC
Consequently, these three models gave the largest and KL even more than 5%. The agreement between
values of torque on the shaft. At OP1, calculated measured and numerical results was the best when the
values of torque were too large when k–ε, k–ε CC KL SST CC KL or SSG RSM models were used, but the
and SSG RSM models were used, while for the other discrepancies were still 4 and 4.4%, respectively.
models the values were very close to the measured
ones. 2.2 Transient Flow Simulation with Different Turbulence
At OP2, turbulence models k–ε, k–ε CC KL Models at One Operating Point
and SSG RSM gave larger values of torque while
We tried to improve the results by transient analysis.
all the other models gave smaller values than the
Three turbulence models were used in this respect:
measurements. At OP3 all models obtained too small
SST, SAS and ZLES. In the case of the SST turbulence
values of torque.
model, the HRS was used for the advection term.
For flow energy losses before the runner, in the
The simulation by the SAS model was done firstly
runner and in the draft tube no measured values were
with BCDS and then also with a HRS, in order to see
available, so only the values obtained by different
the influence of a discretisation scheme. In the case
turbulence models can be compared with each
of the ZLES model, the BCDS was used. For time
other. For all turbulence models the values of losses
discretisation, a second order backward Euler scheme
before the runner, in the runner and in the draft tube
was used.
decreased by using the KL limiter of production term In order to reduce computational time, the domain
and the CC. The KL reduces flow energy losses at for this analysis was without the spiral casing and
stagnation points, therefore its effect was significant stay vane cascade. For the draft tube and draft tube
in stay and guide vane cascades, and to a smaller prolongation, refined grids were used (see Table 1).
degree also in the runner. Curvature correction acted For an appropriate comparison of results the steady-
mostly in the draft tube, where it reduced flow energy state simulation with the SST model was repeated
losses, especially for large discharge. When SSG on the fine grid. In order to see the effect of grid
RSM was used, flow energy losses before the runner density on results, a simulation with ZLES BCDS was
and in the runner were smaller than those obtained by performed on both grids, the basic and the fine one.
two equation models. The difference in losses before The inlet boundary and initial conditions were
the runner obtained by the SST CC KL and SSG prescribed from the steady-state solution of the SST
RSM models was between 0.3 and 0.4% of head at model. The velocity components were prescribed at
all three operating points. The differences in losses the inlet boundary (at guide vane inlet), which means
in the runner obtained by SST CC KL and SSG RSM that the value of discharge was prescribed while the
were at OP1, OP2 and OP3; equal to 0.5, 0.35 and value of head was a result of the simulation. Time
0.31% of head, respectively. The losses in the draft step size corresponded to 0.5 degrees of runner
tube, predicted by the SSG RSM model, were at OP1 revolution. The average value of Courant number was,
slightly larger, at OP2 approximately the same and in all turbine parts, less than 0.3, in the draft tube even
at OP3 significantly larger than those obtained by less than 0.02. Total simulation time corresponded
two-equation models with KL and CC, the difference to 30 runner revolutions. Transient analysis is very
between SST CC KL and the SSG RSM at OP3 was time consuming, therefore it was done only at one
about 0.95% of head. operating point for the blade angle of 28 degrees. This
128 Jošt, D. – Škerlavaj, A. – Lipej, A.
 Strojniški vestnik - Journal of Mechanical Engineering 60(2014)2, 124-134

operating point is not the local best efficiency point. discharge was the input data and it was equal for all
Its guide vane opening and discharge are smaller. The simulations.
numerical analysis had started before the measured
results were available, and the exact positions of the Table 3. The values of viscosity ratio in the draft tube
local best efficiency points had not been known yet. Numerical modeling and Eddy viscosity / Dynamic viscosity
In the case of transient simulations, a certain grid density maximal averaged
amount of time is required for the oscillating values Steady-state SST HRS, FG 10152 2527
of head, torque and efficiency to stabilise around Transient SST HRS, FG 6005 1439
average values. A detailed analysis of the results SAS HRS, FG 1146 169
presented in [2] had showed that turbulent structures SAS BCDS, FG 1311 180
in the flow were after 10 runner revolutions not ZLES CDS, FG 850 81
entirely developed, therefore the simulation time ZLES CDS, BG 1020 95
was extended to 30 runner revolutions. The results
of the last ten runner revolutions were used to obtain The spiral casing with stay vanes was not included
averaged values of torque, head and efficiency. in the computational domain, therefore the flow
In Fig. 3 a comparison of flow in the draft tube energy losses before the runner were calculated as a
obtained by four simulations (steady-state SST, sum of the losses in a guide vane cascade obtained by
transient SST, SAS and ZLES) is presented. In the transient simulation, and the losses in the spiral casing
case of steady-state analysis by SST model there is a obtained by the previous steady-state analysis. It can
large swirl at the end of the cone, which with transient be seen that the losses before the runner are nearly the
simulations nearly disappeared. Streamlines obtained same in all cases.
In the runner, the highest losses were obtained
with the SAS and the ZLES models are more curled
with steady-state, and transient analysis with the SST
than those obtained with SST. For the SAS model, the
HRS (3.81 and 3.93%, respectively) and the smallest
difference in flow due to the discretisation scheme
with SAS BCDS (3.41%) and ZLES BCDS (3.51%).
was hardly visible, therefore only the results of SAS
Losses in the runner obtained by SAS HRS (3.74%)
BCDS are illustrated. For ZLES, a coarser grid in the
are closer to the losses of steady-state and transient
draft tube had a very small influence on streamlines,
SST HRS than to SAS BCDS.
therefore this picture is also omitted. Turbulent
Flow energy losses in the draft tube differ
structure in the flow can be better seen by the iso-
significantly due to steady-state or transient analysis
surface of velocity invariant Q (see Fig. 4), coloured
and due to the choice of a turbulence model and
by viscosity ratio (ratio between eddy and dynamic discretisation scheme. In the case of steady-state
viscosity). With the steady-state SST and also by analysis by SST HRS, the losses in the draft tube
the transient SST model, only large structures in the exceed 6.6% of turbine head, while in case of transient
flow were obtained, but with SAS and especially analysis with the same model they reduce to 3.47%.
ZLES, also small turbulent structures in the flow were The losses obtained with SAS HRS, SAS BCDS and
well resolved. Besides, it can be seen that values of ZLES BCDS are 2.93, 3.0 and 2.94%, respectively.
viscosity ratio are large in the case of the SST model, Steady-state simulation with SST HRS
and much smaller when the SAS and especially the underestimated the value of torque on the shaft by
ZLES models were used (see Table 3). While the 5.07%. With transient simulations the prediction of
influence of discretisation scheme in the case of torque improved significantly. Transient SST HRS
SAS can hardly be seen, the grid density in the case and SAS HRS underestimated values of torque
of ZLES did have an effect on the size of turbulent by 0.78 and 0.71%, respectively. SAS BCDS and
structures and also on the value of viscosity ratio. ZLES BCDS overrated torque by 0.039 and 0.036%,
Small turbulent structures in the flow can be obtained respectively. It seems that for torque prediction a
only on a fine grid where also the values of viscosity choice of differential scheme for advection term was
ratio are smaller. more important than the choice of turbulence model.
In Fig. 5, the results of steady-state analysis (SST When HRS was used, SST and SAS predicted nearly
HRS) and of transient simulations (SST HRS, SAS the same values of torque. Similarly, when BCDS was
HRS, SAS BCDS, ZLES BCDS) are presented. The used, SAS and ZLES predicted nearly the same value.
values of head, torque and efficiency were divided On the other hand, the difference between values
by the measured values. The flow energy losses were obtained with SAS HRS and SAS BCDS is significant
divided by the measured value of head. The value of in spite of the same turbulence model. It is likely

Improvement of Efficiency Prediction for a Kaplan Turbine with Advanced Turbulence Models 129
 Strojniški vestnik - Journal of Mechanical Engineering 60(2014)2, 124-134

Fig. 3. Streamlines and velocity contours in the draft tube; a) steady-state SST, HRS, FG, b) transient SST, HRS, FG, c) SAS, BCDS, FG,
and d) ZLES, BCDS, FG

Fig. 4. Isosurfaces of velocity invariant Q = 0; a) steady-state SST, HRS, FG, b) transient SST, HRS, FG, c) SAS, HRS, FG,
d) SAS, BCDS, FG, e) ZLES, BCDS, FG, and f) ZLES, BCDS, BG

that in the runner the QSAS term was less important of discretisation scheme had a direct effect on the
because close to the runner surface the SAS model predictions of flow close to the runner surface.
acted as the SST model. On the contrary, the choice

130 Jošt, D. – Škerlavaj, A. – Lipej, A.

 Strojniški vestnik - Journal of Mechanical Engineering 60(2014)2, 124-134

a) b)

c) d)
0 = measurement; 1 = steady-state SST, HRS, FG; 2 = transient SST, HRS, FG;
3 = SAS, HRS, FG; 4 = SAS, BCDS, FG; 5 = ZLES, BCDS, FG; 6 = ZLES, BCDS, BG
Fig. 5. Comparison of results obtained with different turbulence models to the measurements of
a) head, b) flow energy losses, c) torque, d) efficiency

The values of calculated efficiency differ operating regimes (Section 2.3) the basic grid was
mostly due to different values of flow energy losses used.
in the draft tube and different values of torque. The
efficiency value calculated from steady-state solution 2.3 Transient Flow Simulation with the Zonal LES Model
with SST HRS was 4.42% smaller than the measured for Different Operating Regimes
one. With transient analysis the results improved
significantly. With SST HRS and SAS HRS the The purpose of this study is to find a turbulence
efficiency values were smaller than the measured ones model that would be capable of predicting efficiency
by 1.01 and 0.24%. The agreement between measured accurately for all operating regimes (OP1, OP2 and
and numerical values is excellent for SAS BCDS OP3). In the previous section, the ZLES model (at
and ZLES BCDS, where the discrepancy is 0.09 and operation regime close to OP3) showed such potential.
0.05%, respectively. Based on the results it can be In order to see whether this model is suitable for all
concluded that the SAS and ZLES models with BCDS operating regimes it should be thoroughly tested at
are very suitable for flow simulation at operating several operating points for three angles of runner
points with large discharge. Results obtained by SAS blades.
HRS are less accurate mostly due to underestimated In this section, simulations with ZLES were
torque. performed for the whole turbine. To reduce
Comparing the results of ZLES BCDS on fine computational time, basic grids for the draft tube and
and basic grids it was seen that the effect of grid draft tube prolongation were used. In all transient
density on the calculated torque, head and efficiency simulations CC and KL are included. The input data
was negligible. Flow energy losses in the draft tube consisted of geometry, head and rotational speed. The
were only higher by 0.18% of head, than on the fine results were the values of discharge, torque on the
grid. Therefore, for transient simulations for different shaft, flow energy losses and efficiency.
Improvement of Efficiency Prediction for a Kaplan Turbine with Advanced Turbulence Models 131
 Strojniški vestnik - Journal of Mechanical Engineering 60(2014)2, 124-134

Fig. 6. Efficiency diagram – measurements and CFD

Table 4. Calculated flow energy losses in the draft tube. ΔH/H×100 operating points with large discharge still remains,
[%] but the improvement with the ZLES model was
Simulation type OP1 OP2 OP3 significant.
Steady-state SST CC KL 1.88 2.29 6.32
Steady-state SSG RSM 2.22 2.43 7.26 3 COMPUTATIONAL EFFORT
SAS SST CC KL ZLES 1.12 1.56 3.33
Transient numerical simulations are very time
The efficiency prediction is presented in Fig. 6. consuming. To get reliable results the grids have to be
All values were divided by the measured efficiency at refined enough and time step must not be too large to
OP2. The diagram shows a clear distinction between get proper values of Courant number. Besides, it takes
the results of steady-state analysis and the results of a long time before the values of efficiency stabilise.
the transient ones obtained with the ZLES model. Usually more than 20 runner revolutions are needed to
With ZLES, the efficiency prediction was improved get stable values.
at all operating points, but the improvement was The simulations were run on a supercomputer
most significant for runner blade angle 28 degrees, cluster with 512 Quad-Core Intel Xeon processors
where the decrease of calculated flow energy losses L5335. For the simulation on the basic grid (8.3
in the draft tube was the largest (see Table 4). At the million nodes) with the ZLES model, 8 quad-core
local best efficiency point for runner blade angle 12 processors (64 cores) were used and the computational
degrees (OP1), the calculated and measured values time was about 28 hours for one runner revolution.
were practically the same. At OP2 (blade angle 20 Too long CPU time is the main disadvantage of
degrees), the efficiency value obtained with ZLES transient simulations and the reason for their limited
was about 0.5% smaller than the measured one. In the use in the design process. It can be expected that with
section 2.2 for blade angle 28 degrees, the discrepancy future development of hardware and software the
in efficiency was only 0.05%. That operating point problem will be overcome.
corresponded to the third point from the left on the
ZLES curve for 28 degrees In Fig. 6 the efficiency 4 CONCLUSIONS
values at the first three points on the ZLES curve
for blade angle 28 degrees agreed with the measured • Steady-state analysis failed entirely to predict
results well. At higher discharge, the discrepancy flow in the Kaplan turbine. Even for small and
increased. Peak to peak difference in efficiency values optimal runner blade angles, where efficiency
was 0.44% but numerically obtained position of the was quite accurately predicted, a detailed analysis
local best efficiency point was shifted to the left. The of results showed that the prediction of torque
tendency that numerical prediction is less accurate at and discharge values (head as input data) was
132 Jošt, D. – Škerlavaj, A. – Lipej, A.
 Strojniški vestnik - Journal of Mechanical Engineering 60(2014)2, 124-134

not accurate enough. At full rate the efficiency 6 NOMENCLATURE

prediction was extremely poor. The discrepancy
was more than 4%, mostly due to underestimated cr1, cr2, cr3 [-] Empirical constants for frot
torque on the shaft and overestimated flow energy frot [-] Curvature correction function
losses in the draft tube. The results were only g [m·s–2] Gravitation
slightly improved by using the KL and CC. H [m] Head
• Transient simulations with SST HRS, SAS HRS, HExp [m] Head, experimental value
SAS BCDS and ZLES BCDS were performed at ΔH [m] Flow energy losses
one operating point for a maximal runner blade ΔH = ΔPtot/(ρg)
angle. The results were significantly improved. k [m2·s–2] Turbulence kinetic energy per
The largest improvement was achieved with SAS unit mass
BCDS and ZLES BCDS, where not only the L [m] RANS length scale
efficiency values but also the values of torque and LvK [m] Von Karman length scale
head (discharge as input data) were almost equal M [Nm] Torque on the shaft
to the measured values. MExp [Nm] Torque on the shaft,
• Comparing transient results of SST HRS, SAS experimental value
HRS and SAS BCDS, it can be concluded that the Ptot [Nm–2] Total pressure
improvement due to the use of BCDS instead of Q [m3 s–1] Discharge
HRS, was even larger than the improvement due QSAS [kg·m–3·s–2] Source term in ω-equation for
to the use of SAS instead of SST. With BCDS, the SAS SST model
the agreement with measurements was improved r [m] Runner radius
r , r*
mostly because of smaller losses in the runner [-] Non-dimensional arguments
in frot
and better prediction of torque on the shaft.
• The effect of grid refinement in the draft tube,
β [degrees] Runner blade angle
on efficiency prediction with ZLES BCDS, was
ε [m2 s–3] Turbulence dissipation rate
negligible in spite of clearly seen differences in
η [-] Efficiency
vortex structures. A grid with about 2 million
ηExp [-] Efficiency, experimental value
nodes was in this case fine enough for reliable
ηrel [-] Efficiency divided by
results.
measured efficiency at OP2
• It is important that transient simulations are not
φ [-] Discharge coefficient
stopped too soon, otherwise swirls from the φ = Q / (πωr3)
steady-state simulations used as initial conditions ρ [kg m–3] Density
still remain, and also vortex structures are not ψ [-] Energy coefficient
entirely developed. ψ = 2gH/(ωr)2
• Simulations with ZLES BCDS were performed ω [s–1] Runner speed
at several operating points for three runner blade ω [s–1] Turbulence frequency
angles. The discrepancy in efficiency values was
smaller than 1% at all operating points. It can BCDS Bounded central differential scheme
be concluded that ZLES model with BCDS is BEP Best efficiency point
suitable for all operating regimes. BG Basic grid
CC Curvature correction
5 ACKNOWLEDGEMENTS FG Fine grid
HRS High resolution scheme
The authors are grateful to their colleagues and KL Kato-Launder limiter of production term in
especially to the head of the Turbine R&D Department, equation for turbulent kinetic energy
Dr. Vesko Djelić, for geometrical data and measured SAS Scale adaptive simulation
results. The research was partially funded by the SST Shear stress transport
Slovenian Research Agency ARRS – Contracts No. ZLES Zonal large eddy simulation
P2-0196 and 1000-09-160263.

Improvement of Efficiency Prediction for a Kaplan Turbine with Advanced Turbulence Models 133
 Strojniški vestnik - Journal of Mechanical Engineering 60(2014)2, 124-134

7 REFERENCES an invariant dynamical systems approach. Journal

of Fluid Mechanics, vol. 277, no. 1, p. 245-272,
[1] Motycak, L., Skotak, A., Obrovsky, J. (2010). Analysis DOI:10.1017/S0022112091000101.
of the Kaplan draft tube effect. IOP Conference Series: [15] Egorov, Y., Menter, F. (2008). Development and
Materials Science and Engineering, vol. 12, no. 1, p. application of SST-SAS turbulence model in the
012038, DOI:10.1088/1755-1315/12/1/012038. DESIDER project. Peng, S.-H., Haase, W. (eds.),
[2] Jošt, D., Škerlavaj, A., Lipej, A. (2012). Numerical flow Advances in Hybrid RANS-LES Modelling. Springer,
simulation and efficiency prediction for axial turbines Heidelberg, p. 261-270, DOI:10.1007/978-3-540-
by advanced turbulence models. IOP Conference 77815-8_27.
Series: Materials Science and Engineering, vol. 15, no. [16] Fröhlich, J., Terzi, D. (2008). Hybrid LES/RANS
6, p. 062016, DOI:10.1088/1755-1315/15/6/062016. methods for the simulation of turbulent flows. Progress
[3] Gebart, B.R., Gustavsson, L.H., Karlsson, R.I. (2000). in Aerospace Sciences, vol. 44, no. 1, p. 349-377,
Proceedings of Turbine-99 – Workshop on draft tube DOI:10.1016/j.paerosci.2008.05.001.
flow. Luleå University of Technology, Luleå. [17] Menter, F.R., Gabaruk, A., Smirnov, P., Cokljat, D.,
[4] Cervantes, M.J., Engström, T.F., Gustavsson, L.H. Mathey, F. (2010). Scale-Adaptive Simulation with
(2005). Turbine-99 III. Luleå University of Technology, Artificial Forcing. Peng, S.-H., Doerffer, P., Haase,
Luleå. W. (eds.), Progress in Hybrid RANS-LES Modelling.
[5] Marjavaara, D., Kamakoti, R., Lundstöm, T.S., Springer, Berlin, p. 235-246, DOI:10.1007/978-3-642-
Siddharth, T., Wright, J., Schy, W. (2005). Steady and 14168-3_20.
unsteady CFD simulations of the Turbine-99 draft tube [18] Menter, F., Egorov, Y. (2009). Formulation of the
CFX-5 and STREAM. Cervantes, M.J., Engström, Scale-Adaptive Simulation (SAS) Model during the
T.F., Gustavsson, L.H. (eds.), Turbine-99 III. Luleå DESIDER Project. Haase, W., Braza, M., Revell,
University of Technology, Luleå, p. 83-100. A. (eds.), DESider - A European Effort on Hybrid
[6] Kurosawa, S., Nakamura, K. (2005). Unsteady RANS-LES Modelling. Springer, Berlin, p. 51-62,
turbulent flow simulation in Turbine-99 draft tube. DOI:10.1007/978-3-540-92773-0.
Cervantes, M.J., Engström, T.F., Gustavsson, L.H. [19] Adamian, D., Travin, A. (2011). An efficient generator
(eds.), Turbine-99 III. Luleå University of Technology, of synthetic turbulence at RANS–LES interface in
Luleå, p. 73-82.
embedded LES of wall-bounded and free shear flows.
[7] Cervantes, M.J., Andersson, U., Lövgren, H.M. (2010).
Kuzmin, A. (ed.), Computational Fluid Dynamics
Turbine-99 unsteady simulations - Validation. IOP
2010. Springer, Berlin, p. 739-744, DOI:10.1007/978-
Conference Series: Materials Science and Engineering,
3-642-17884-9.
vol. 12, no. 1, p. 012014, DOI:10.1088/1755-
[20] Smirnov, P.E., Menter, F. (2009). Sensitization of the
1315/12/1/012014.
SST turbulence model to rotation and curvature by
[8] Nilsson, H., Cervantes, M.J. (2012). Effects of inlet
applying the Spalart-Shur correction term. Journal
boundary conditions on the computed flow in the
of Turbomachinery, vol. 131, no. 4, p. 041010,
Turbine-99 draft tube, using OpenFOAM and CFX. IOP
DOI:10.1115/1.3070573.
Conference Series: Materials Science and Engineering,
vol. 15, no. 3, p. 032002, DOI:10.1088/1755- [21] Kato, M., Launder, B.E. (1993). The modelling of
1315/15/3/032002. turbulent flow around stationary and vibrating square
[9] Jošt, D., Drešar, P. (2011). Numerical Analysis of the cylinders. Proceedings of the 9th Symposium on
Flow in an Axial Turbine by Different Turbulence Turbulent Shear Flows, p. 10.4.1-10.4.6.
Models, Report Nr. 3046. Turboinštitut, Ljubljana (in [22] Barth, T.J., Jespersen, D.C. (1989). The design and
Slovene). application of upwind schemes on unstructured
[10] ANSYS (2012). ANSYS CFX-Solver Theory Guide. meshes. 27th Aerospace Sciences Meeting, AIAA Paper
Ansys, Canonsburg. 89-0366.
[11] Wilcox, D.C. (1994). Turbulence Modelling for CFD. [23] Darwish, M.S., Moukalled, F. (2003). TVD schemes
DCW Industries, La Cañada. for unstructured grids. International Journal of Heat
[12] Menter, F.R. (1994). Two-equation eddy-viscosity and Mass Transfer, vol. 46, no. 4, p. 599-611, DOI:
turbulence models for engineering applications. 10.1016/S0017-9310(02)00330-7.
AIAA Journal, vol. 32, no. 8, p. 1598-1605, [24] Jasak, H., Weller, H.G., Gosman, A.D. (1999).
DOI:10.2514/3.12149. High resolution NVD differencing scheme for
[13] Menter, F.R., Kuntz, M., Langtry, R. (2003). Ten arbitrarily unstructured meshes. International
years of industrial experience with the SST turbulence Journal for Numerical Methods in Fluids, vol.
model. Hanjalić, K., Nagano, Y., Tummers, M. (eds.), 31, no. 2, p. 431-449, DOI:10.1002/(SICI)1097-
Turbulence, Heat and Mass Transfer 4. Begell House, 0363(19990930)31:2<431::AID-FLD884>3.0.CO;2-T.
New York, p. 625-632. [25] IEC 60193 (1999). Hydraulic turbines, storage
[14] Speziale, C.G., Sarkar, S., Gatski, T.B. (1991). pumps and pump-turbine – Model acceptance tests.
Modelling the pressure-strain correlation of turbulence: International Electrotechnical Commission, Geneva.

134 Jošt, D. – Škerlavaj, A. – Lipej, A.